A space contracting particle swarm optimization and its application in investment prediction

Abstract

Multi-layer feed-forward network has a good ability of function approximation, but the usual training algorithm, BP algorithm may easily fall into local minimum and it has weak generalization ability. While the space contraction particle swarm optimization (SCPSO) algorithm has a good capability of global search, the training algorithm for multi-layer feed-forward network is constructed on the basis of the SCPSO. Considering the nonlinear feature of the investment issue, a multi-layer feed-forward network model is established. The SCPSO algorithm as learning algorithm is applied to training of multi-layer feed-ward network and then a simulated prediction is made. The comparison of the prediction result between the network based on SCPSO and BP network indicates that the former has high prediction accuracy.